Fwd: Integrating speech speed control

[Bill Cox]

On Thu, Nov 4, 2010 at 9:12 AM, Christopher Brannon
<chris at the-brannons.com> wrote:
> Bill Cox <waywardgeek at gmail.com> writes:
>> So, if you have 33 milliseconds or more of input sound data, it can
>> make progress
>
> Yes, this could be a nice add-on for Speech Dispatcher, even though I
> personally can't understand speech at such high rates. ?My limit is
> somewhere between 500 and 600 WPM, I believe.
> Does understanding require lots of practice?

Hi, Chris.  Understanding does in fact require practice, especially at
my age, begin 47.  I may never reach really high speeds like Sina (he
listens at 6X or higher), but I'm pretty happy listening to voxin at
about 3.2X speed up.  What I've done for about the last year and a
half is listen to high speed audio books on my Android phone, which I
create with the voxin "say" program, though I had to modify it a bit.
After every book, I increased the speed just a bit.  I listen now at
-s113 most often.

I personally can't understand espeak very well at 3X speed up, but I
believe that it is simply a matter of practice, a it was with voxin.
The native espeak speed up capability is hampered by at least two
limitations in the algorithm.  First, it synthesizes entire pitch
periods, and cannot deal with fractional pitch delays.  This causes
silence, for example before both T's in the word "test", to be overly
long at -s450.  Second, some sounds in espeak are played from .wav
files, and these are not sped up.  This is why some consonants sound
overly long at -s450.  To compensate for these delays, pauses at the
end of phrases are mostly eliminated at -s450, further hampering the
quality.  Using Sonic to speed up espeak gets around these
limitations, and I find the result more understandable and pleasing.

I'm also listening, for the first time, to an audio book read by real
people, from librivox.org.  The default speed is very irritating, but
I'm enjoying it at 2X.  I really want a new Android application that
allows me to dynamically vary the speed, the way I can when using Orca
or speakup.

Let me know if I can help with integration.  I think a good place
would be in the audio back-end, just before speech samples are sent to
the audio drivers.  We can also help test it over in Vinux, and if
it's done soon, we may get it into the Vinux release based on
Maverick.

Bill